Hidden biodiversity in Neotropical streams: DNA barcoding uncovers high endemicity of freshwater macroinvertebrates at small spatial scales

Results We collected approximately 300 specimens across the four sites; however, our analysis focused on the 224 individuals that were successfully barcoded (Table 1 and S1 Table). We were able to identify nearly 70% of individuals to genus level using the morphological approach, but spe- cies-level identification was only possible for 56 individuals (25%). Some of the most numer- ous taxa across sites included Leptophlebiidae (11.2% of individuals), Libellulidae (11.2%), Naucoridae (6.7%), Notonectidae (6.3%), Chironomidae (5.4%), Gerridae (4.9%), Hydropsy- chidae (4.0%), Perlidae (3.6%), and Baetidae (3.6%). The 224 COI sequences revealed a total of 168 haplotypes (S1 Table). After comparing these haplotypes with molecular data from Mu´rria et al [25], we found 153 (~91%) unique haplo- types from the Panama Canal Watershed. In Frijolito, the site sampled by the two studies, we found a total of 46 haplotypes, 45 of which were unique to our study. Our final COI dataset consisted of 171 terminals, including the 168 new barcode sequences, and 3 outgroup sequences retrieved from GenBank. The final aligned and pruned dataset con- tained 620 aligned positions, including gaps, with 371 variable sites, of which 359 were parsi- mony-informative (~96% of variable positions). As expected, we observed a hierarchical increase in the mean K2P genetic divergence with increasing taxonomic levels from within a species 0.38% (SE = 0.002), to within family 9.92% (SE = 0.01), to within order 19.75% (SE = 0.01). However, we were not able to identify our specimens to species level from our

BLAST search, given that only around 50% of our sequences matched existing data in the pub- lic databases, and most of these matches corresponded to genus and family level only. Both ML and BI inference trees for all specimens showed well-defined clades at the level of order and family, with some differences in the topology, but overall support was higher for the BI tree, which we used to represent the number of molecular species (S1 Fig and S2 Fig). Our species delimitation analyses yielded variable, but relatively high numbers of species. Specifically, GMYC detected 106 MOTUs (95% confidence intervals: 104–109), whereas ABGD found a total of 100 MOTUs (S1 Fig). These ABGD results were confirmed indepen- dently of the chosen model (Jukes-Cantor and Kimura) and were unaffected by changes of prior limits for intraspecific variation and threshold. When looking at spatial patterns of diversity, we observed some overlap in the number of shared MOTUs as well as a considerable number of unique haplotypes in each river: Frijoles (65), Frijolito (43), Trinidad (31) and Indio (22) (Fig 2). This pattern was also supported by the Fisher’s alpha diversity index, which showed variation in molecular species among sites: Frijoles 53.40, Frijolito 22.80, Trinidad 17.01, and Indio 67.63. Whittaker’s index of β diversity also showed high species turn-over across sites (0.87). Similarly, our rarefaction/extrapolation analyses showed variation in species richness across sites: Frijoles (52), Frijolito (27), Trinidad

(22) and Indio (22). However, the most striking pattern was a lack of saturation in the Table 1. Macroinvertebrate lineage diversity in the Panama Canal Watershed. Site Ind. Hap. GMYC ABGD Trinidad (TRI) 45 33 22 21 Frijolito (FTO) 65 46 27 26 Frijoles (FES) 88 72 52 50 Indio (IND) 26 24 22 21 All 224 168 106 100 The data show the number of successfully barcoded individuals (Ind.), cox1 haplotypes (Hap.), and molecular species (based on GMYC and ABGD) collected at four sites. https://doi.org/10.1371/journal.pone.0231683.t001 PLOS ONE Hidden biodiversity in Neotropical streams PLOS ONE | https://doi.org/10.1371/journal.pone.0231683 August 7, 2020 6 / 13